This part of ISO24614 presents the basic concepts and general principles of word segmentation, and provides language-independent guidelines to enable written texts to be segmented, in a reliable and reproducible manner, into word segmentation units (WSU).
NOTE1 In language-related research and industry, the word is a fundamental and necessary concept. It is thus critical to have a universal definition of what comprises a word for the purposes of segmenting a text into words. One cannot simply use rules based only on spaces and punctuation to delimit words. Such rules do not account for situations such as hyphenated compounds, abbreviations, idioms or word-like expressions that contain symbols or numbers. Word segmentation is even more problematic for languages that do not use spaces to separate words, such as Chinese and Japanese, and for agglutinative languages, where some functional word classes are realized as affixes, such as Korean.
The many applications and fields that need to segment texts into words — and thus to which this part of ISO24614 can be applied — include the following.
Translation
Word count is the principal method for calculating the cost of a translation. Word segmentation is a standard function in translation memory systems and computer-assisted translation (CAT) tools. Word segmentation is performed by term extraction tools, which are sometimes provided in terminology management systems and CAT tools.
Content management
Most content management systems and databases allow for searching by individual words. The content being searched has to be segmented to permit matching with a search word. Furthermore, search functions require knowledge of the boundaries of words.
Speech technologies
Text-to-speech systems generate speech based on words and therefore require word segmentation for lexicon lookup, stress assignment, prosodic pattern assignment, etc.
Computational linguistics
Various natural language processing (NLP) systems must segment text into words in order to carry out their functions. NLP systems include
-
morphosyntactic processors,
-
syntactic parsers,
-
spellcheckers,
-
text classification systems, and
-
corpus linguistics annotators.
Lexicography
Lexical resources are often evaluated by size, usually by referring to the number of words.
NOTE2 The size of language resources is an essential benchmark for their management. Quantifying the size of language resources is typically achieved by counting the words. However, because NLP applications use different segmentation methods, each calculates the number of words differently and arrives at a different sum for the same text. A reliable, reproducible, standard measure would allow comparable results. This is not to say that applications may not use their own, application-specific segmentation methods. For example, a speech synthesis application might segment a text into smaller or larger units compared to another application.